Journal of Economic Cooperation 24, 1 (2003) 1-24
WAGE DIFFERENCES BY GENDER, WAGE AND SELF
EMPLOYMENT IN URBAN TURKEY
Yusuf Ziya Özcan*, Şenay Üçdoğruk** and Kıvılcım Metin Özcan***
This paper explains wage differences by gender, wage and self employment in
an urban setting in Turkey. Data employed is taken from the 1994 Household
Income Survey of the State Institute of Statistics (SIS) of Turkey. The Oaxaca
decomposition of the wages into discrimination and endowment components
indicates the existence of a relatively higher discrimination in the wage
employment than in the self employment. In the context of returns to
education, self employment provides the highest returns for men in self
employment. This shows that education is highly valued in the self
employment than in the wage employment.
I. INTRODUCTION
Studies to determine wage differentials have evoked considerable
interest in both developing and industrialised countries. Depending upon
the characteristics of their labour markets, factors producing wage
differentials in those studies varied from race, gender, education, job
status, occupation, type of sector (public vs private) to type of industry,
among the others.
A glance at the literature reveals that there exist two main
approaches to study wage differentials. The earlier approach, the human
capital model, explains wage differences (and their decline over time) by
the relative educational attainment and quality of education obtained by
individuals (Becker, 1964; Mincer, 1958; Mincer, 1974). It is implied
that observed wage differences are due to different educational
*
Associate Professor, Middle East Technical University, Department of Sociology,
Ankara, Turkey.
**
Associate Professor, Dokuz Eylul University, Department of Econometrics, Buca,
Izmir, Turkey.
***
Corresponding author. Associate Professor, Bilkent University, Department of
Economics, Ankara, Turkey.
2
Journal of Economic Cooperation
attainments of individuals. A more recent approach stresses
discrimination in wage determination and tries to decompose wage
differences into two components: discrimination, which includes
differential evaluation of individuals’ characteristics in the labour
market, and endowments of individuals (Oaxaca, 1973). Inclusion of
both endowment, which is analogous to human capital model, and
discrimination, which is known to exist in almost all labour markets, can
be considered an improvement over the human capital model. This paper
adopts the discrimination approach and tries to explain wage differences
by job status in an urban setting in Turkey.
Turkey offers a unique opportunity to study wage differentials due
to the characteristics of its labour force and prevailing income
inequalities. Tansel (1999: 2) describes the Turkish labour market’s
characteristics as “high rates of population and labour force growth,
declining rates of participation and exceptionally low levels of female
participation in urban areas.” Turkey has a very segregated labour
market where women are traditionally employed in agriculture. In
1970, 89.5 % of women compared to 54% of men were employed in
agriculture. Despite the restructuring of the Turkish economy over the
past 20 years, which brought a reduction in the agricultural workforce,
the overwhelming majority of Turkish women are still employed in
agriculture. In urban areas, the women’s participation rate is 16%. Of
these, 10% are unpaid family workers. On average, women receive
50% of male earnings which points out a considerable discrepancy in
wages –a discrepancy observed for each occupation, job status and
level of schooling. In rural areas, the segregation is even more marked.
90% of working women are employed in agriculture and 97% of them
work as unpaid family workers. In industry and services, women’s
participation remains as low as 18.4% and 14%, respectively
(Dayıoğlu, 1995: 3).
Due to considerable variation in its labour market, Istanbul is
selected as a study site for the analysis of this article. High labour
participation of women in Istanbul provides a chance to observe wage
differences due to many of the variables of interest, particularly job
status. Data employed is taken from the 1994 Household Income
Survey of the State Institute of Statistics (SIS) of Turkey. The wage
estimation model developed combines two sets of variables. The first
set includes variables tapping endowments of individuals, which
Wage Differences in Urban Turkey
3
makes up, in a sense, an enhanced human capital model. The second
set is composed of work-related variables. The two sets are combined
in the model without any à priori restriction. Due to the cross-sectional
nature of the data employed, homoscedasticity assumption is tested.
Sample selectivity bias (or the inverse Mill’s ratio) for each job status
plus gender category is carried out by using the two-step Heckman
procedure (Heckman, 1979). The Oaxaca (1973) decomposition is used
to decompose the wages into discrimination and endowment
components.
The organisation of the paper is as follows: Section II summarises
the studies on wage discrimination and wage decomposition in
developed and developing countries as well as Turkey. Section III
defines the data and variables of interest of the wage equation. Section
IV estimates an empirical wage model, and reports the econometric
results for men and women in wage and self employments, respectively.
Section V presents the Oaxaca decomposition of wages with respect to
the discrimination and endowment components. The last section
presents the concluding remarks.
II. LITERATURE ON WAGE DISCRIMINATION
Since the work of Oaxaca, wage determination has been subject to
scrutiny in terms of its components. Oaxaca, in the introduction of his
article, notes that “Culture, tradition and overt discrimination tend to
make restrictive the terms by which women participate in the labour
force. These influences combine to generate an unfavourable
occupational distribution of female workers and to create pay
differences between males and females.” (Oaxaca, 1973: 693). In fact,
there are two types of discrimination at work in developing countries,
defined as pro- and post- discrimination. Pro-discrimination occurs
when certain groups are prevented from joining the labour market. In the
case of women, this is obvious since they are not sent to school as much
as males, so their chance to participate in the labour force is reduced
considerably. Post-discrimination operates within the labour market.
Employers, under the effects of the cultural characteristics of society,
evaluate endowments of individuals differently. It is considered
appropriate for women to work in certain occupations, mostly lower
paying jobs (Cohen, 1971). Even in the same occupation there are
certain levels that they are not allowed to attain.
4
Journal of Economic Cooperation
It is clear that post-discrimination is more pronounced in developed
countries where gaps producing pro-discrimination among various
groups tend to close as in the U.S., where the gap between the
educational attainments of blacks and whites tend to disappear. Unlike
developed countries, developing countries are plagued with both proand post-discrimination. Mean years of schooling for males by region in
Turkey vary from a highest of 6.48 for Marmara to a lowest of 4.75 in
Southeast Anatolia. Corresponding figures for females are 5.76 and
1.81, respectively (Tansel and Güngör, 1992: 8). The discrepancy is due
to attitudes toward education of daughters and differences among the
regions in terms of economic development and hence the availability of
educational facilities.
Decomposition of wage is helpful to see what we call postdiscrimination in the labour market and has been used extensively in
wage differential research in both developing and developed countries1.
Due to the importance and high visibility of male-female wage
differences, most of these works deal with decomposition of wage for
gender which is followed by race and ethnicity. Only a small portion of
the above studies deal with other determinants of income such as full
time/part time working in Manning and Robinson’s (1998: 389) article
and public-private sector employment in Tansel’s (1999) work. With
few exceptions, most of the studies used cross-sectional data for one
year (Ashraf, 1994; Ashraf and Ashraf, 1998; Darity et. al., 1995).
While only Fairlie (1999) uses panel data for 1968-1989 the Manning
and Robinson (1998) study employs limited panel data for Britain.
The application of wage decomposition highlighted the issue of
controlling the sample selectivity bias (Heckman, 1979) which is
considerably important for some labour markets. In labour markets
which are segregated and discriminatory, selecting a representative
group of disadvantaged individuals (women, blacks, public employees)
becomes problematic and OLS estimates will be biased. Due to the low
level of participation (high pro-discrimination) of women in the
workforce, sample selectivity bias is considered more appropriate for
women than men. However, some studies argue that sample selectivity
bias is a more serious problem for the male population (Winer and
Gindy, 1992; Arends, 1992). It appears that sample selectivity bias
should also be controlled for men especially in countries where the
unemployment rate is relatively high (Dayıoğlu, 1995). It is also true
Wage Differences in Urban Turkey
5
that in a number of studies the coefficient of sample selectivity term was
found to be insignificant (Scott, 1992; Tenjo, 1992; Yang, 1992). Such
conclusions render the selectivity problem an empirical question rather
than something to be known à priori (Dayıoğlu, 1995).
Katz (1997) investigated the gender gap in a Russian industrial town.
Based on previous studies done by others, she reports a serious income
gap between males and females in Russia, which is 2/3 in favour of
males. She estimated log-linear wage equations separately for men and
women for both hourly and monthly wages. Sample selectivity bias was
not controlled since she was sure that sampled women were
representative of others at large. Decomposition of the gender wage
differential showed that “given the Soviet wage-structure which was in
itself male-biased, differences in experience, education, qualification
level and work conditions account for roughly one-third of the
differences in hourly wages. Despite the broad range of factors
controlled, Soviet women were paid less because they were women. A
bias is built into perceptions of productivity.” (Katz, 1997: 446).
In one of the first studies on wage decomposition in developing
countries, Ashraf and Ashraf (1993) find a high level of discrimination
against women in Rawalpindi City. Years later, Ashraf and Ashraf (1998)
report for Pakistan a substantial decline in the gender earnings gap
between 1979 and 1985-86 which is valid for four provinces and across
every industrial group. Similarly, Kingdon (1997) finds for India rising
rates of returns to education by education level, yet girls face significantly
lower economic rates of returns to education than boys. In a study on
Malaysian data, Nor (1998) suggests that the gender gap, which is
determined by productivity characteristics, present family variables and
occupational distribution of women and men is largely due to labour
market discrimination, which is a result of a favouritism towards men
rather than an unfavourable female treatment in the labour market.
In her study of gender income inequalities, Dayıoğlu (1995: 200)
highlights earnings variations not only by gender but also by sector,
region, occupation and educational attainment. Vast differences in
income by these variables are documented. By adopting the
discrimination approach, she indicates two sources for gender income
inequalities: 1) Human capital differences between genders and 2)
Different valuation of the productive characteristics of the two groups,
6
Journal of Economic Cooperation
that is, discrimination in the labour market. By employing the Oaxaca
technique, she carries out a decomposition of both male-female and
region wage equations. She concludes that “Regardless of the type of
decomposition, the most important factor contributing to the earnings
gap between genders is determined to be the different valuations of
individual productive traits or discrimination against women in the
labour market”(Dayıoğlu, 1995: 221).
Tansel (1996) investigates self and wage employment of men and
women with respect to residential segregation (rural vs urban), age and
education, after providing profiles of the wage- and self-employed in
Turkey. Her particular interest is to find determinants of employment
choice and estimation of wages for the wage- and self-employed. She
observes that the “fraction of self-employed declines in favour of wage
earners during the process of development.”(Tansel, 1996: 21). This
movement suggests that mobility may be switching from self
employment to wage employment. She finds that education has a greater
effect on wage employment choice of women than of men. Women’s
earnings are more responsive to education than men’s for the wage
earners. In a similar way, education seems to determine self-employed
men’s earnings more strongly than the wage earners. The present study,
in some ways, details Tansel’s work.
More in line with discrimination literature is Tansel’s (1999) work on
public-private employment choice and wage differentials of the two
groups. Controlling for observed characteristics and sample selection,
public administration wages are found to be at parity or lower than covered
private sector, in particular at the university level. Wages at state-owned
enterprises are higher than covered private sector wages except at the
university level. While the public administration has a more egalitarian
wage structure, the covered private sector exhibits a large gender gap in
wages, which suggests a stronger discrimination in the private sector.
Comparing covered and uncovered earners, she notes that the first have
two and a half times higher wages than the latter (Tansel, 1999).
Motivated by previous studies, this paper aims at contributing to the
existing empirical literature by estimating an empirical wage model for
men and women in wage and self employment and decomposing the
wages into discrimination and endowment components in urban Turkey
in the following subsections.
Wage Differences in Urban Turkey
7
III. DATA AND VARIABLES
The data are obtained from the 1994 Household Income Survey of the
State Institute of Statistics of Turkey (SIS). Instead of using the whole
data set, which covers 19 cities, the decision has been made to use only
the data for Istanbul on the grounds that it is the largest city in Turkey
which offers a chance to study wages of individuals in all job statuses
and income strata. Yet, only the individuals who are working in nonagricultural jobs and between the ages of 15 and 65 are included in the
analysis. With these restrictions, sample size declined to 1324.
The data set includes the determinants of the earnings differentials of
each working individual. Specifically, information on gender, age,
education, marital status, occupation, economic activity (industry),
public-private employment, number of workers at the workplace, the
number of hours worked per week, social security coverage and job
experience is employed. Only the net activity income of each individual
is used in the analysis. Net activity income, then, includes wages and
salaries, entrepreneurs’ income, income earned from manufacturing and
construction sectors and mining activities, and service and trade income,
after taxes and social security payments.
By definition, unpaid family workers (UFW) do not have any net
activity income. Although a few reported in kind income, no estimation
is attempted for the unpaid family workers, which reduces the number of
categories of job status variable to four: wage-salary earners, casual
workers, employers and the self-employed. Furthermore, for the
estimation of wage equations, a two-category job status variable is used
where 1 represents wage employment (wage-salary and casual workers)
and 2 represents self employment (employers and the self-employed).
Information on occupation is not used and is replaced by industry in
dummy format since there is a considerable overlap between the two and
the industries have never been employed in the previous research. Job
experience is calculated in the same way as described by Oaxaca, that is
EXP=AGE-EDUC-6. Due to a high correlation between age (AGE) and
experience (EXP), age is not used except for descriptive purposes
(Oaxaca, 1973). Despite the existence of detailed information about the
kind of social security coverage, a binomial coverage variable is created
and turned into dummies. A similar approach is followed for a five-
8
Journal of Economic Cooperation
category marital status variable, which is expressed as a binomial (1 for
married and 2 for others). Two kinds of education scale are available for
the analysis. The first captures the level of schooling by school (or
diploma) type completed. The second is created by the researchers by
assigning years corresponding to each level. Whenever necessary,
dummies representing each category of education variable are created
and used particularly in wage estimation equations. Private-public
employment separation is expressed in dummy format.
A. Job status
The labour market structure of Turkey resembles more to that of
developed countries where wage employment is relatively higher than self
employment compared to the developing countries where self
employment is relatively higher. In 1955, 43.8 percent of the male labour
force was self-employed as opposed to 20.5 percent wage earners. In
1990, self employment declined to 30.7 percent while wage employment
rose to 50.1 percent. Tansel (1996) reports similar trends for the Republic
of Korea, Taiwan and China. In the same period, increases both in self
employment and wage employment of women are observed due to the
increased participation of women in the labour force.
Table 1: Employment Composition of
Urban Population and Istanbul by Gender (%)
Employment Status
Wage and Salary Earners
Casual Workers
Employers
Self-employed
Unpaid Family Workers
Turkey (Urban)
Men
Women
56.6
64.2
9.6
6.5
12.6
1.1
16.5
13.0
4.2
15.2
Men
56.2
15.6
9.2
15.0
4.0
Istanbul
Women
77.2
8.4
2.4
6.0
6.0
A closer look at the distribution of job status in urban Turkey and
Istanbul reveals that an even higher proportion of the urban labour force
is in wage employment. Table 1 summarises percentages of people in
different job statuses by gender.
If the hypothesis that the higher the proportion of wage employment,
the higher the economic development is tested with these data, Turkey,
and particularly Istanbul, seems to have adopted a similar path as the
Wage Differences in Urban Turkey
9
developed countries. The percentage of self-employed men for Istanbul
is just a few points above the percentage for Germany (8.9%).
B. Income
In the job market of Istanbul, men have higher incomes compared to
women. This is true regardless of the job status held. The highest mean
income for men belongs to employers, followed by the self-employed,
wage-salary earners and casual workers. The employers, on average,
make 3.2 times more than the self-employed, 5.4 times more than the
wage-salary earners and 7 times more than the casual workers. Mean
male earnings indicate tremendous differences. This is also true for
women, although to a lesser degree. Unlike men, wage-salary earner
women make, on average, more than the self-employed. Again,
employer women make 5.5 times of the casual and 2.7 times of the
wage-salary earners. Similar sizable differences are also observed
between men and women. Expressed as a percentage of men’s income,
women’s income is 37 for employers, 25 for the self-employed, 75 for
wage-salary earners and 47 for casual workers. It is interesting to find
the highest earnings discrepancy in the self-employed and the least
discrepancy in the wage-salary group. Table 2 compares income,
education and age of the labour force in Istanbul by job status.
Table 2: Means and Standard Deviation of Some Independent
Variables by Job Status and Gender
Inc*
Log Inc
Educ
Age
Exp
Hours
Wage-Salary
M
W
95.41
71.6
101.82
87.9
18.05
17.70
.78
.87
7.36
8.26
3.57
4.14
32.91
28.80
10.74
9.41
19.55
14.54
11.05
10.98
53.19
47.13
14.34
10.87
Casual
M
W
74.0
34.5
53.4
27.5
17.93
16.96
.64
1.04
5.52
4.90
2.03
4.50
32.24
34.67
10.60
12.37
20.72
23.76
11.30
15.04
56.58
38.19
13.96
24.50
Employer
M
W
519.7
191.5
776.0
200.4
19.53
18.54
1.01
1.30
8.48
10.67
4.22
2.58
39.35
36.17
9.75
10.89
24.87
19.50
10.41
12.63
53.72
53.50
14.98
37.45
Self-employed
M
W
162.4
40.3
200.7
48.9
18.48
16.78
1.01
1.39
6.51
6.67
3.30
4.30
40.54
40.00
11.08
10.51
28.03
27.33
11.49
12.58
55.65
27.53
19.88
15.42
UFW
M
W
-
-
-
-
7.64
3.33
26.39
11.30
12.75
12.22
54.32
15.21
6.27
3.47
39.00
9.29
26.73
9.51
47.28
20.88
W: women, M: men, UFW: unpaid family worker, Inc: income, Log Inc: logarithm of
income, Educ: education, Exp: experience, Hours: weekly working hours.
* Million Turkish Liras.
1
The mean.
2
The standard deviation.
10
Journal of Economic Cooperation
C. Education
In a study that employs national data, wage-salary earners are found to
have the highest education compared to other groups (Tansel, 1996: 8).
In Istanbul, employers seem to have the highest education followed by
the wage-salary earners. Given the economic potential of Istanbul,
which attracts the most educated and qualified people, this is no
surprise. What is more interesting is the higher educational attainments
of women among the employers, wage-salary earners and selfemployed. Higher education of the employers may be an indication of a
trend from traditional to schooled businessmen. Higher women’s
participation in the employer group may mean that setting up and
running a business requires relatively higher education and women have
to be even better to establish a business in the male-dominated labour
market.
D. Age, experience and weekly working hours
Two variables, age and experience, show a similar pattern due to the way
the experience variable is calculated. Since experience is more relevant to
determination of wages, interpretation here will be limited to experience
only. The self employment group (employer and self-employees) have
higher mean experience compared to other status groups. This is
explained by the nature of the private work that takes a longer time to
build and, traditionally, the lack or limited application of pension or
retirement funds in self-employed jobs. It is interesting to note that
women who work as casual workers and unpaid family workers have
higher mean years for experience. This again can be explained by the
absence of social security provisions, particularly for the unpaid family
workers who are forced to work much longer in order to survive.
The average number of hours worked per week is higher for men
(53.62) than for women (45.36) in Istanbul. The casual workers put in
more hours (56.58) than any other group, followed by the self-employed
(55.65), unpaid family workers (54.32), employers (53.72) and wagesalary earners (53.19). Order by work hours per week changes for
women in that employer women work the most per week (53.50)
followed by women in unpaid family work (47.27), women in the wagesalary group (47.13), women in casual work (38.19) and women in self
employment (27.53).
Wage Differences in Urban Turkey
11
E. Industry affiliation and public vs. private employment
The industry that the individual works in has not been dealt with in wage
determination studies in Turkey. Instead, occupations are used in the
efforts to determine wage (Metin and Üçdoğruk, 1997). For Istanbul and
its very divided labour market, the idea of using and trying industries
was appealing. In fact, for Istanbul, industry is found to be more
correlated with income than occupation. Originally, industries are coded
into 9 main categories, but due to the small numbers in the first two
categories and the electricity, gas and water category, they are regrouped
into six categories as manufacturing, construction and maintenance,
trade (wholesale and retail), transport, communication and storage,
finance and services (societal and personal). Mean wages in industries
exhibit substantial differences in that transport has the highest income
(204.7 m. TL) followed by trade (162.5 m. TL), finance (124.4 m. TL),
services (116 m. TL), construction (102.7 m. TL) and manufacturing
(93.6 m. TL).
Table 3: Cross-tabulation of Industry by Job Status and
Employment Type
Industry
Wage Emp.
Manufacture
44.3
Construction
9.7
Trade
20.5
Transport
5.6
Finance
1.9
Services
17.9
χ2=87.7 d.f.=10 p=.001
Job Status
Self Emp.
26.0
11.2
38.2
12.6
0.4
11.6
χ2=415.2
Employment Type
UFW
Public
Private
28.8
9.9
43.2
16.9
-11.6
42.4
2.8
27.9
5.1
10.6
6.7
1.7
2.8
1.4
5.1
73.8
9.2
d.f.=5 p=.001
Total
39.7
10.4
25.3
7.1
1.6
16.0
In order to see the job status of workers in industries, the two
variables are cross-tabulated. The manufacturing industry employs at
least one fourth of the workers in each status group and is highly
represented among wage-salary earners (44 %). Unpaid family workers
are highest in the construction industry, which is followed by the selfemployed. Trade is the primary industry for self-employed and unpaid
family workers, while only one-fifth of wage-salary earners work in
trade. The higher representation of unpaid family workers in trade is
explained by the higher frequency of family-owned businesses in
12
Journal of Economic Cooperation
Istanbul. The self-employed make the modal category in the transport
sector where the other status groups are represented with 5 percent each.
While there are relatively more wage-salary earners in the services
industry, all three status groups are represented rather weakly in the
finance sector.
F. Marital status
Marital status is considered as one of the determinants of wage. It is
speculated that responsibilities to take care of a family increase the
worker’s motivation to find higher paying jobs and work longer hours to
meet the needs of a family. In general, married workers earn more than
unmarried ones in Istanbul, where the mean earning for the married is
148.5 million TL compared to 63.2 million TL for the unmarried. This
difference varies according to the job status of the individuals.
Unmarried ones earn on average 54.2 percent of the married in the
wage-salary group, 83.3 percent in the casual worker group and 52
percent in the employer group. Among the self-employed, the difference
is negligible (also insignificant) and in favour of the unmarried (t=1.445,
df=177, p=.150).
G. Social security coverage
It is generally accepted that the effect of social security coverage on
wages is not direct, and on average, minimal. Therefore, coverage in any
form is not expected to generate considerable income differentials.
However, the income gap is sizable between covered and uncovered
workers in Turkey. In the case of Istanbul, the mean income for covered
workers is higher than uncovered ones (143.8 m vs. 94.5 m). Moreover,
within the covered and uncovered groups, there exist considerable
differences by gender. Among the covered workers, the mean income
for men is 156.9 m while it is 88.6 m for women. While uncovered men
make 108.9 m, uncovered women make 28.3 m. Of course, the observed
differences in income cannot be explained by the existence or absence of
the coverage. Cross-tabulation of coverage status with education clearly
indicates that coverage itself is related to education (χ2=103.1, df=5,
p=.001). It is logical to think that more educated workers have higher
inclination toward being covered on their jobs.
Wage Differences in Urban Turkey
13
IV. DETERMINATION OF THE INCOME DIFFERENTIALS
The model used in the following sections includes two sets of variables.
The first is known as human capital variables that tap qualities of
individuals such as education and experience. Variables in the second
set relate to work and the individual’s position at work (see Note 2).
A. Estimates for Men in Wage and Self Employment
The estimates of the log earnings equation for wage earner men in Table
4 indicate that the majority of the categories of education–the most
frequently used human capital variable–are significant with expected
signs. The illiterates and primary school graduates do not seem to be
gaining any returns for their education as compared with the workers
who are literate but with no schooling (base category). This is different
for junior high school, high school and university and more educated
wage earners who make consistently higher returns to their higher
educational attainments.
The negative sign of the social security coverage is expected,
implying that having social security shelter actually lessens wage
earners’ income as compared to those who are not covered (base
category). It is a fact that security coverage is not spread in Turkey.
Paying insurance premiums and accepting other responsibilities born
out of insurance is considered as a burden by the employer. Instead of
being responsible to the State due to insured workers, many employers
prefer to pay some extra wages to workers who do not demand
insurance.
As the significance of the dummies for industry indicate, being in
manufacturing, construction and finance industries does not bring
workers any more income than the wage earners in the services industry
(base category). However, the wage workers in the trade and transport
industries gain 35 percent and 42 percent more than the service workers.
Work hours per week have a small positive effect on log income of
wage earners. Being married means 29 percent more income over the
income of the unmarried wage earner men, which is in line with general
expectations. Similarly, job experience has a positive effect on the log
income of the wage earner men.
14
Journal of Economic Cooperation
For self-employed men, education is more profitable since returns to
education are higher compared to returns to education in wage
employment. Self-employed men gain much more for the same education
in self employment than their counterparts in wage employment. This
may mean a higher valuation of education in the self employment. As
opposed to the trend in wage employment, workers who are covered make
little more than the uncovered in self employment.
Table 4: Selectivity Corrected Estimates of Wage Equations for Men
Variables
Wage Employment
Coefficient
P > |z|
16.2191
0.001
Self Employment
Coefficient
P > |z|
15.8027
0.001
Constant
Education
Illiterate
-0.2243
0.246
-0.2772
0.733
PS
0.1934
0.207
0.7544
0.162
JHS
0.3403
0.033
1.0739
0.082
HS
0.7518
0.001
1.2447
0.017
UNIV+
1.6124
0.001
1.7273
0.001
Socsec Cov
-0.0800
0.099
0.0199
0.919
Industry
Manufacture
0.0916
0.166
0.4220
0.097
Construction
0.0763
0.417
0.3402
0.244
Trade
0.3542
0.002
0.4930
0.049
Transport
0.4160
0.001
0.5887
0.042
Finance
0.0134
0.948
0.8609
0.448
Hours
0.0050
0.001
0.0086
0.023
Married
0.2987
0.002
0.1531
0.631
Experience
0.0908
0.001
0.0668
0.015
Experience2
-0.0015
0.001
-0.0012
0.012
Selection term
-0.5873
0.0315
0.2546
0.35871
R2
0.3981
0.1706
F test
31.71
0.0001
3.18
0.0001
N
784
264
1
Standard error.
PS: primary school. JHS: Junior high school and equivalent. HS: high school and
equivalent. UNIV+: university and above. Socsec Cov: social security covered.
Hours: weekly working hours.
Experience2: experience squared.
N: number of observations.
A similar case to education is true for the industries. Workers in
trade and transport make little more over the service industry workers
than their counterparts in the wage employment. Hours worked per week
Wage Differences in Urban Turkey
15
have again negligible contribution to the income of self-employed men,
which is little over what their counterparts make in wage employment.
Being married contributes positively to the income of self-employed
men over the unmarried, although this is a little less than what marriage
brings to men in wage employment. The same interpretation is true for
experience. Each year of experience increases the income of the selfemployed men, which is just under the corresponding gain in wage
employment.
B. Estimates for Women in Wage and Self Employment
Estimates for women in wage employment in comparison with wage
earning men indicate some major differences between the sexes. Returns
to high school and university and higher education are higher for
women. Illiterate women earn less than women who are literate without
diploma.
As opposed to wage earner men, having social security shelter
increases women’s income over the uncovered ones. This might be due
to the much lower income of women, whose insurance premiums do not
bring too much financial burden on the employer.
None of the dummies for the industry variable are significant. This
can be interpreted to mean that no matter which industry women work
in, there is no additional contribution stemming just from working in a
particular industry.
Unlike wage earner men, the more hours worked per week do not
translate into higher income for women. As pointed out earlier, women
work on the average fewer hours than men, which can be explained, in
part, by the traditional responsibilities imposed on women in connection
with the domestic work.
As expected, married wage earner women make less than their
unmarried counterparts. Factors explaining this phenomenon vary from
household responsibilities imposed on women to interruptions due to
child bearing. Although job experience increases income of women, this
increase is much less than what wage earner men gain from one
additional year on the job.
16
Journal of Economic Cooperation
Estimates for the women in self employment are unreliable due to
the small number of self-employed women. Therefore, no interpretation
will be advanced for them.
Table 5. Selectivity Corrected Estimates
of Wage Equations for Women
Variables
Wage Employment
Coefficient
P > |z|
16.7780
0.001
Constant
Education
Illiterate
-1.4784
PS
0.2329
JHS
0.2208
HS
0.7945
UNIV+
1.70484
Socsec Cov
0.4140
Industry
Manufacture
-0.0611
Construction
0.2731
Trade
-0.1112
Transport
-Finance
-Hours
-0.0002
Married
-0.2711
Experience
0.0286
Experience2
0.0005
Selection term
-0.5873
R2
0.4337
F test
9.90
N
196
1
Standard error.
N: Number of observations.
Self Employment
Coefficient
P > |z|
9.3407
0.001
0.001
0.494
0.552
0.027
0.001
0.001
0.8387
-2.5104
-7.0875
-1.7450
-0.5361
1.7008
0.790
0.032
0.001
0.267
0.795
0.077
0.638
0.596
0.505
--0.955
0.042
0.092
0.238
0.03151
1.2362
9.4030
3.2329
--0.0334
0.2386
-0.0044
-0.0012
0.9555
0.6192
3.32
21
0.289
0.001
0.030
--0.052
0.706
0.024
0.097
0.06001
0.0001
0.074
V. DECOMPOSITION OF WAGES BY JOB STATUS
Wage decomposition by way of the Oaxaca technique and its various
derivatives has been used extensively in studies on wage differentials
and determination, some of which are mentioned in the section dealing
with the discrimination literature. The Oaxaca decomposition is a
technique which is very useful in distinguishing the difference in wages
due to human capital characteristics (i.e. education, age, experience),
from those that could be attributed to discrimination (i.e. the
unexplained portion of the difference in wages).
Wage Differences in Urban Turkey
17
Table 6 summarises the calculations performed for both wage and
self employment by taking the male and female structure separately as a
base for each employment type. This was necessary due to the index
number problem inherent in the technique that makes no distinction
between the male and female structures. So, both are employed in the
analysis. Additionally, all calculations performed on the wage equations
are estimated by OLS rather than the selectivity corrected wage
equations on two grounds: in half of the wage equations, the sample
selectivity bias is found insignificant, and, more importantly, the
purpose here is to see the actual degree of discrimination prevalent in
the labour market, which does not impose the use of sample selectivity
corrected estimates. The OLS wage equations used for this exercise are
different than the ones used in the previous section, due to the skewed
Table 6. Wage Decomposition by Job Status
Male1
Female2
Employment
Total
Due to
Due to
Due to
Due to
Variables
Type
Difference Discrimination Endowment discrimination Endowment
W
Illiterate
.0412
.0103
.0309
.0091
.0320
A
PS
.0147
-.0053
.0200
.0273
-.0126
G
JHS
.0300
.0062
.0238
.0368
-.0068
E
HS
.0026
-.0379
.0405
.0259
-.0233
UNIV+
-.0562
-.0743
.0181
.0095
-.0657
E
Socsec Cov
-.4087
-.0007
-.4080
-.4017
-.0070
A
Industry
-.0094
-.0005
-.0089
-.0077
-.0017
R
Hours
-.0613
.0358
-.0971
-.1116
.0503
N
Married
.2014
.0473
.1541
.2412
-.0398
I
Experience
1.4530
.3715
1.0815
1.3867
.0663
N
Experience2
-.8602
-2.2568
-.6033
-.8762
.0161
G
Total
.4320
.0955
.3365
.4244
.0076
S
E
L
F
E
M
P
L
O
Y
M
E
N
T
1
2
Illiterate
PS
JHS
HS
UNIV+
Socsec Cov
Industry
Hours
Married
Experience
Experience2
.1062
1.1104
.3351
.8961
.5640
-1.2100
-.0799
-.6592
-.4636
-1.1278
.0140
.0093
.1111
-.0118
-.1013
-.0135
.0068
.0079
-.3062
.0012
.1021
-.0646
.0969
.9993
.3469
.9974
.5775
-1.2168
-.0879
-.3530
-.4648
-1.2299
.0786
.0385
1.4176
.3035
.5818
.5363
-1.3436
-.0439
-.0055
-.6521
-1.3147
.0845
.0677
-.3073
.0316
.3143
.0277
.1336
-.0360
-.6536
.1885
.1868
-.0704
Total
1.6942
-.2591
1.9533
1.8114
-.1172
A + (-) sign indicates advantage for males (females).
A + (-) sign indicates advantage for females (males).
18
Journal of Economic Cooperation
distribution of women in the categories of INDUSTRY and
WORKPLACE variables. For industry, the dummies are dropped and
workplace variable is not included.
It appears that wage differentials (in logarithmic forms) for both
employment sectors and sexes are considerable, which is in agreement
with the earlier findings (Dayıoğlu, 1995; Tansel, 1996; Tansel, 1999).
Gender difference in wage employment seems substantially smaller
than it is in self employment (.4320 vs. 1.6942). When the male
structure is taken as a base, the proportion of discrimination in the
gender wage difference is 22% in favour of men. When the female
structure is used, this climbs to 99 percent in favour of women. Similar
comparison for self employment indicates the seriousness of
discrimination in wages in Istanbul (15% of the wage difference when
the male structure is used and 107% when the female structure is
used). Besides pointing out the sizable wage discrimination in self
employment–nearly three times of that in the wage employment–the
sign of discrimination term suggests female advantage even when the
male structure is used. This runs against both the academic consensus
and the layman’s perception about women’s earnings in Turkey.
Moreover, there is enough indication that this is an artifact of the
model used that is under-specified for securing comparable estimates
for males and females in each sector. Oaxaca (1973: 699) himself
explains this clearly by saying that “the magnitude of the estimated
effects of discrimination crucially depends upon the choice of control
variables for the wage regressions.” Therefore, the interpretation
should be limited to wage employment, for which wage equations
indicate a better fit.
VI. CONCLUSION
This paper investigates the factors that determine the wages of both men
and women in wage employment and self employment with the view
that wages are determined rather differently in both employment types.
It is pointed out that for wage and self-employed workers, Turkey
resembles industrialised countries more than developing ones in terms of
division of labour force. This confirms the claim of development
theorists who assert that with development, the proportion of the selfemployed declines in the labour force.
Wage Differences in Urban Turkey
19
Istanbul is chosen to represent the urban population in Turkey.
Having a population of approximately 7 million and being the heart of
the Turkish economy, Istanbul is the ideal urban setting to investigate
processes that shape wages.
The model used is composed of both human capital variables and
work-related variables. Using the two-step Heckman procedure, it is
shown that there are differences in the wage determination of men and
women and between wage and self employment. The important
difference between wage employment and self employment is the higher
returns to education in self employment. This indicates the existence of
different valuation of education in the two employments. For men in
both employment types and women in wage employment, more
education means more income. This walks in the face of those who
downplay the role of education in income generation. In recent years,
due to increased favouritism, particularly in the public sector job
placements, some have started believing that education is of little help in
job competitions. This paper disputes such thinking and indicates the
increasing importance of education.
Women in wage employment seem to benefit most from social
security coverage while social security coverage appears as a
disadvantage for men in wage employment. As explained earlier, this is
due to the tradeoff between high wage but no coverage and low wage
but coverage.
Industries that workers are employed in did not confirm the pattern
expected by the researchers. Only the workers in the trade and transport
sectors seem to benefit, compared to workers in the services industry,
from their respective industries. It is known that Istanbul is the center for
wholesale and retail trade. Therefore, trade’s contribution to income
more than the others is not surprising. It is tempting to think that
contribution of the transport industry is just a spillover effect of the trade
activities. Since the highest volume of exports and imports are handled
in Istanbul, which requires colossal transportation facilities of all kinds
(land, sea and air), workers benefit more from such industry where
wages are second to the wages in transport.
Working hours per week appear to be contributing to men’s earnings
while more hours imply reduction of women’s wages. Women usually
20
Journal of Economic Cooperation
work less compared to men. Cultural responsibilities for domestic
housework and child-rearing activities constitute the biggest obstacles
that prevent women from working longer and without disruption. This is
also confirmed by the effect of marriage on earnings. While men in both
wage and self employment clearly benefit from marriage compared to
the unmarried, being married seems to be a disadvantage for women in
wage employment.
Job experience contributes positively to men in both employments
and to women in wage employment. However, an extra year of job
experience brings relatively more income to wage earner men than to
men in self employment, and brings the least to women in wage
employment.
Decomposition of the wages into discrimination and endowment
components indicates the existence of a relatively higher discrimination
in wage employment than in self employment. This may seem
paradoxical given the fact that income differentials are higher in self
employment than in wage employment. This apparent paradox resolves
when different valuation of the two employments is taken into account.
As shown above, in the context of returns to education, self employment
provides highest returns for men in self employment. This clearly shows
that education is more highly valued in self employment than in wage
employment. Furthermore, since education is an important determinant
of earnings, workers being paid on the basis of their education level goes
along with the idea of income according to endowment, which means
less discrimination. All these suggest that whatever difference remains
in wages is more due to endowments than discrimination in self
employment. In wage employment, different criteria for evaluating
workers seem to be utilised, which are known to be more prone to
favouritism. Moreover, relatively less income differentials are due to
restrictions imposed by laws regulating salaries in the case of public
employment and efforts of unions in the case of private employment,
which also affect non-union workers in some indirect ways.
NOTES
1. Ashraf and Ashraf (1993; 1998) applied wage decomposition for
Pakistan, Kingdon (1997) for India, Nor (1998) for Peninsular
Malaysia, Katz (1997) for a Russian industrial town, Appleton,
Wage Differences in Urban Turkey
21
Hoddinott and Krishnan (1996) for three African countries (Ethiopia,
Uganda and Côte d’Ivoire), Drago (1989) for Australia, Ashraf
(1994), Darity, Guilkey and Winfrey (1995) and Fairlie (1999) for
the U.S., and Dayıoğlu (1995) and Tansel (1996; 1999) for Turkey.
2. The model employed in this paper combines the two sets without
imposing any restriction on them. That is, it is assumed that all the
variables included in the model influence wage. The model, then,
includes education, work experience and marital status, which are
commonly used in the human capital models, as well as social
security coverage, public-private sector employment, industry and
the number of hours worked per week. In the model, education,
industry, social security coverage, employment type and marital
status are expressed as dummy variables.
REFERENCES
Appleton, S., Hoddinott, J., and Krishnan, P., (1996), “The Gender
Wage Gap in Three African Countries”, Centre for the Study of African
Economies Working Paper Series 96/07.
Arends, M., (1992), “Female Labor Force Participation and Earnings in
Guatemala”, in Women’s Employment and Pay in Latin America, ed. G.
Psachharapolous and Z. Tzannatos, World Bank, Washington, D.C.
Ashraf, J., (1994), “Recent Trends in White-Black
Differentials”, Economic Letters, 45, 487-493.
Earnings
Ashraf, J. and Ashraf, B., (1993), ”Estimating the Gender Wage Gap in
Rawalpindi City”, Journal of Development Studies, 29, 365-376.
Ashraf, J. and Ashraf, B., (1998), “Gender, Earnings and Discrimination
in Pakistan” Indian Journal of Labour Economics, 41, 317—326.
Becker, G.S., (1964), Human Capital, Columbia University Press, New
York.
Cohen, M. S., (1971), “Sex Differences in Compensation”, Journal of
Human Resources, 434-447.
22
Journal of Economic Cooperation
Darity, W. Jr., Guilkey, D. K. and W. Winfrey, (1995), “Ethnicity, Race,
and Earnings”, Economic Letters, 47, 401-408.
Darity, W. Jr., Guilkey, D. K. and W. Winfrey, (1996), “Explaining
Differences in Economic Performance among Racial and Ethnic Groups
in the USA: The Data Examined”, American Journal of Economics and
Sociology, 55, 411-425.
Dayıoğlu, M., (1995), “Earnings Inequality Between Genders in
Turkey”, Unpublished PhD Thesis, The Middle East Technical
University, Ankara, Turkey.
Drago, R., (1989), “The Extent of Wage Discrimination in Australia”,
Australian Bulletin of Labour, 15, 313-325.
Fairlie, R.W., (1999), “The Absence of the African-American Owned
Business: An Analysis of the Dynamics of Self-Employment”. Journal
of Labor Economics, 17, 80-108.
Heckman J.(1979), “Sample Selection Bias as a Specification Error”.
Econometrica, 47(1),153-161.
Katz, K., (1997), “Gender, Wages and Discrimination in the USSR: A
Study of a Russian Industrial Town”, Cambridge Journal of Economics,
21, 431-452.
Kingdon, G.G., (1997), “Does Labor Market Explain Lower Female
Schooling in India”, STICERD, London School of Economics,
Discussion Papers Series DERP/1.
Manning, A. and Robinson, H., (1998), “Something in the Way She
Moves: A Fresh Look at an Old Gap”, London School of Economics,
Centre for Economic Performance, Discussion Paper no: 389.
Metin, K. and Üçdoğruk, Ş., (1997), “Đstanbul Đlinde Gelir Farklılıklarını
Belirleyen Etmenler: Đnsan Sermayesi Modeli (1994)”, Ekonomik
Yaklaşım, 8, 283-302.
Mincer, J., (1958), “Investment In Human Capital And Personal Income
Distribution”, Journal of Political Economy, 66, 281-302.
Wage Differences in Urban Turkey
23
Mincer, J., (1974), Schooling and Earnings, New York, Columbia
University Press.
Nor, L. M., (1998), “An Overview of Gender Earnings Differentials in
Peninsular Malaysia”, IIUM Journal of Economics and Management, 6,
23-49.
Oaxaca, R., (1973), “Male-female Wage Differentials in Urban Labor
Markets”, International Economic Review, 14, 693-709.
Oaxaca, R.L. and Ransom, M., (1994), “On Discrimination and the
Decomposition of Wage Differentials”, Journal of Econometrics, 61, 521.
Oaxaca, R.L. and Ransom, M., (1998), “Calculation of Approximate
Variances for Wage Decomposition Differentials”, Journal of Economic
and Social Measurement, 24, 55-61.
Scott, K., (1992), “Women in the Labor Force in Bolivia” in Women’s
Employment and Pay in Latin America, ed. G. Psachharapolous and Z.
Tzannatos, World Bank, Washington, D.C.
Tansel, A., (1996), “Self Employment, Wage Employment and Returns
to Education for Urban Men and Women in Turkey”, in Education
Labor Market in Turkey, ed. Tuncer Bulutay, State Institute of Statistics,
Ankara, Turkey.
Tansel, A., (1999), “Public-Private Employment Choice, Wage
Differentials and Gender in Turkey”, Economic Growth Center Yale
University, Discussion Paper no: 797.
Tansel, A., and Güngör, N.D., (1997), “The Educational Attainment of
Turkey’s Labor Force: A Comparison Across Provinces and Over
Time”, METU Studies in Development, 24, 531-547.
Tenjo, J., (1992), “Labor Markets, the Wage Gap and Wages in
Colombia”, in Women’s Employment and Pay in Latin America, ed. G.
Psachharapolous and Z. Tzannatos, World Bank, Washington, D.C.
24
Journal of Economic Cooperation
Winer, C. and Gindy, T. H., (1992), “Women’s Labor Force
Participation and Earnings in Honduras”, in Women’s Employment and
Pay in Latin America, ed. G. Psachharapolous and Z. Tzannatos, World
Bank, Washington, D.C.
Yang, H., (1992), “Female Labor Force Participation and Earnings
Differentials in Costa Rica”, in Women’s Employment and Pay in Latin
America, ed. G. Psachharapolous and Z. Tzannatos, World Bank,
Washington, D.C.